Related Experiment Videos
Stereoscopic depth: its relation to image segmentation, grouping, and the recognition of occluded objects
K Nakayama1, S Shimojo, G H Silverman
1Smith-Kettlewell Eye Research Institute, San Francisco, CA 94115.
Perception
|January 1, 1989
Summary
Depth encoding is key for distinguishing object contours. This research shows that understanding depth helps visual systems separate intrinsic object boundaries from extrinsic occlusion boundaries, improving object recognition.
Area of Science:
- Computer Vision
- Computational Neuroscience
- Perception
Background:
- Objects in images are defined by intrinsic contours (object shape) and extrinsic contours (occlusion).
- Extrinsic contours can hinder object recognition and require early differentiation from intrinsic contours.
- Depth perception is hypothesized to be critical for distinguishing these contour types.
Purpose of the Study:
- To investigate the role of depth encoding in differentiating intrinsic and extrinsic contours.
- To determine if depth information aids in image segmentation and object recognition.
Main Methods:
- Analysis of contour properties related to object shape versus occlusion.
- Hypothesis testing on the critical role of depth encoding in contour discrimination.
- Experimental validation using stereoscopic depth planes for object recognition tasks.
Main Results:
- Common borders are perceived as intrinsic to closer objects and extrinsic to farther objects.
- Intrinsic borders aid image segmentation, while extrinsic borders facilitate grouping.
- Object recognition improved when partially occluded objects were presented in a front stereoscopic depth plane compared to a back plane.
Conclusions:
- Depth encoding is a crucial mechanism for distinguishing intrinsic and extrinsic contours in visual processing.
- This distinction is essential for accurate surface description and object recognition.
- The findings support a model where depth perception influences how visual information is segmented and grouped.